What We Learned Hosting a Finance Breakfast in Prague
Earlier this year we started asking a simple question to finance leaders we met at events, on calls, and in roundtables: where do you actually start with AI in finance?
The answers were consistent enough that we decided to do something about it. We invited 30 CFOs, finance directors, and finance managers to a business breakfast at our Prague office. A morning with peers who are all trying to answer the same question: where do I actually start?
Who was in the room
CFOs from manufacturing, retail, logistics, hospitality, real estate, and financial services. Companies ranged from 20-person startups to Prague Airport. Some were existing Keboola customers. Most were not. What they had in common: they all run finance teams that are doing more with less, and they're all trying to figure out where AI fits.
Before the morning, we called almost everyone individually. We asked three questions. The answers were more specific than we expected.
Where finance teams are with their data today
Half the room described themselves as still getting a sense of what's out there. No active project, no vendor evaluation, just trying to understand the landscape. Eight were actively evaluating options and planning a change within the next year. Three were ready to move within three months.
The split is important: most finance leaders are not behind because they lack ambition. They're behind because there is no obvious starting point.
What they need most in the next 3 to 6 months
We asked everyone to pick their priorities. Two answers dominated:
Start utilizing AI for financial workflows and reduce manual work to create more strategic capacity. These two came up in almost every call. They're different problems on the surface but connected underneath. The manual work (data gathering, reconciliation, report assembly) is exactly what prevents the strategic work. And AI is the tool people believe can bridge that gap, even if they don't know how yet.
Improving real-time visibility came third. Then increasing data quality, accelerating the close cycle, and enabling self-service analytics.
Where they are on the AI journey
This one surprised us.
12 people were already building their own automations using tools like Claude Code, Cursor, or similar AI assistants. They're not waiting for IT to roll something out. They're experimenting on their own, often with personal accounts, sometimes without official approval.
11 had some automation in place and were actively exploring AI use cases. 12 were still mostly Excel-driven. A few were exploring built-in copilots like Microsoft Copilot or Gemini.
There is no "standard" AI maturity in mid-market finance. In the same room, you had a CFO who uses Claude to prototype cash forecasts and a finance manager who still manually consolidates reports from five subsidiaries in Excel.
And several are in the middle of ERP decisions - some migrating to a new system and trying to figure out what to build on top before the team disbands. Others were questioning whether a full migration is still the right move. All of them know that ERP alone is never enough - you always end up building something on top of it.
Five conversations that stuck
A few moments from the individual calls kept coming back during the event.
The solo experimenter. A CFO who is the only person on his finance team working with AI. He already has Claude connected to their data models and uses it regularly. He asked it about year-end cash position. The AI didn't know whether to calculate through the budgets or how to handle data stored at different granularities. The result was about 30% off. He's not discouraged. He understands the problem is in how the data is structured, not in the AI itself. He wants to learn how to adjust the data foundation so the answers get better. His goal: automate financial planning, budgeting, and cash forecasting so his team stops spending so much time just looking for data.
The ERP window. A CFO mid-migration to a new ERP system, thinking about what to build on top before the implementation window closes. She knows that once the ERP is live and the project team disbands, building anything new becomes ten times harder. The question is what's worth building now versus later.
The hiring question. A CEO calculating whether hiring a controller is still the right answer, or whether the right platform could make that role unnecessary. Not because they don't value people, but because they're running a small team and every hire needs to cover more ground than just one function.
The data quality problem. A CFO whose biggest concern is not tools or AI but whether the accounting data underneath is correct. "Which tool connects to it doesn't really matter. The question is how to guarantee the data is right." He's running Abra Gen flowing into Metabase, and the reports aren't fully trustworthy because the source data has quality issues.
The "not yet kissed by AI" CFO. His words, not ours. He wants to keep up with the times, came specifically to hear from our customers like Creditinfo that are already doing it, and can already imagine AI handling reporting, forecasting, and closing processes. He just needs to see it working in practice first.
Different stages. The same underlying question: where do I start, and how do I know the data is ready?
What was said on stage
A huge thank you to Jakub Žalio, Group CTO of CreditInfo, and Štěpán Pittauer for joining Pavel on stage and speaking openly about what transformation actually involves, including the parts that did not go as planned. That kind of honesty is rare and it made the conversation worth having.
Michal Hruška showed what a governed metric catalog looks like in practice, with real data, live. The strongest feedback came from the people who came specifically for that.
The morning reinforced what we keep hearing: the challenge in finance AI is not ambition. It is the data underneath. Most teams are working with data that is fragmented, manually assembled, and not ready to hand to AI without first building a proper foundation. And the hardest part, as one attendee described, is not the technology. It is getting the whole organisation mentally ready for what a solid foundation unlocks.
The conversation is not unique to Prague
We have had versions of it with finance leaders in London, in US roundtables, and across dozens of individual calls. The questions are consistent. Where do we start? How do we know when the data is ready? What does AI in finance actually look like in practice?
If you are asking the same questions, there are two conversations worth listening to.
In Episode 1 of our podcast series Journey to AI-Ready Finance, Jiří Maňas and Jakub Žalio talk about why AI keeps failing in finance and what needs to be in place first.
In Episode 2 (coming up), Thilo Kusch, Group CFO at P3 Logistic Parks, and Jiří talk through what it actually took to build a data foundation across 14 systems and 11 countries, and what changed once it was in place.
